I’ll be blunt. I’ve spent months building a product on Gemini 3 Flash. After testing 3.5 Flash for one day, I’m genuinely scared for my business.
Let me be clear about what happened here: You took a model that could do PhD-level math reasoning at $3/1M output tokens. You “upgraded” it into a model that CANNOT do the same math, removed temperature control so developers can’t even tune it, lowered the default thinking from high to medium without warning, and then charged 3x more for it. And then you published a blog post calling it “frontier intelligence that scales with you.” It doesn’t scale with me. It broke my product. You hid the GPQA and AIME benchmarks from the announcement. We both know why. If the numbers were good, they’d be on the front page of the blog post. The most frustrating part: your own migration guide says “For Computer Use workloads, continue using Gemini 3 Flash Preview.” So YOU know 3.5 Flash isn’t a complete replacement. But you won’t say whether 3 Flash will stay alive. Developers are not asking for faster coding agents. We’re asking you not to destroy reasoning capability that took you a trillion parameters to build, then pretend it didn’t happen. Questions: 1. Why were GPQA and AIME scores hidden? 2. Will 3 Flash Preview be deprecated? Give us a date or confirm it’s staying. 3. Is math reasoning restoration on the 3.5 Flash roadmap? 4. Who at Google decided that a $0.50/$3.00 reasoning tier no longer needs to exist? I have Korean high school students depending on this model to analyze their exam performance. There is no alternative. Not Claude, not GPT, not open source. Only Gemini 3 Flash. Don’t pull the rug.